{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:SNTZPZRFUXY7GWP6URUOSH65FC","short_pith_number":"pith:SNTZPZRF","schema_version":"1.0","canonical_sha256":"936797e625a5f1f359fea468e91fdd288a8425d64f592a45763738d80095d5c4","source":{"kind":"arxiv","id":"2608.02064","version":1},"attestation_state":"computed","paper":{"title":"Geometry-Guided Layerwise FFN Width Allocation in Transformers","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Mikhail Burtsev, Radu State, Timur Mudarisov","submitted_at":"2026-08-03T11:08:58Z","abstract_excerpt":"Feed-forward networks (FFNs) account for a large fraction of Transformer parameters, yet their hidden width is usually constant across depth. We ask whether this capacity can instead be allocated from a forward-pass measurement of layer behavior. We view each FFN as transporting a cloud of token representations and quantify the induced geometric change using correspondence-preserving shift, Gromov-Wasserstein distortion, and degree-one persistent homology under raw and scale-normalized metrics. A layerwise approximation surrogate yields an exact fixed-budget optimizer. Across seven pretrained "},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":false,"formal_links_present":false},"canonical_record":{"source":{"id":"2608.02064","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-08-03T11:08:58Z","cross_cats_sorted":["cs.AI","cs.CL"],"title_canon_sha256":"3db78f4d08d5970801e5787e838d80fc7c589fc403a11f001230a4b0d32f959e","abstract_canon_sha256":"dbebe50bfe786b70db0b4fc8364a4cd549d4373f6a87ba4893f491b264b068e8"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-08-04T02:10:46.342911Z","signature_b64":"/ybMl6mJKcZNQEraaRVJsism6MxHnqyVsMAhEJe20inrkGQlpyJP455M0AOwRrkoIQAZziELCHIQdozHjO9nCg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"936797e625a5f1f359fea468e91fdd288a8425d64f592a45763738d80095d5c4","last_reissued_at":"2026-08-04T02:10:46.341408Z","signature_status":"signed_v1","first_computed_at":"2026-08-04T02:10:46.341408Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Geometry-Guided Layerwise FFN Width Allocation in Transformers","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Mikhail Burtsev, Radu State, Timur Mudarisov","submitted_at":"2026-08-03T11:08:58Z","abstract_excerpt":"Feed-forward networks (FFNs) account for a large fraction of Transformer parameters, yet their hidden width is usually constant across depth. We ask whether this capacity can instead be allocated from a forward-pass measurement of layer behavior. We view each FFN as transporting a cloud of token representations and quantify the induced geometric change using correspondence-preserving shift, Gromov-Wasserstein distortion, and degree-one persistent homology under raw and scale-normalized metrics. A layerwise approximation surrogate yields an exact fixed-budget optimizer. Across seven pretrained "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2608.02064","kind":"arxiv","version":1},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2608.02064/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"aliases":[{"alias_kind":"arxiv","alias_value":"2608.02064","created_at":"2026-08-04T02:10:46.342950+00:00"},{"alias_kind":"arxiv_version","alias_value":"2608.02064v1","created_at":"2026-08-04T02:10:46.342950+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2608.02064","created_at":"2026-08-04T02:10:46.342950+00:00"},{"alias_kind":"pith_short_12","alias_value":"SNTZPZRFUXY7","created_at":"2026-08-04T02:10:46.342950+00:00"},{"alias_kind":"pith_short_16","alias_value":"SNTZPZRFUXY7GWP6","created_at":"2026-08-04T02:10:46.342950+00:00"},{"alias_kind":"pith_short_8","alias_value":"SNTZPZRF","created_at":"2026-08-04T02:10:46.342950+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/SNTZPZRFUXY7GWP6URUOSH65FC","json":"https://pith.science/pith/SNTZPZRFUXY7GWP6URUOSH65FC.json","graph_json":"https://pith.science/api/pith-number/SNTZPZRFUXY7GWP6URUOSH65FC/graph.json","events_json":"https://pith.science/api/pith-number/SNTZPZRFUXY7GWP6URUOSH65FC/events.json","paper":"https://pith.science/paper/SNTZPZRF"},"agent_actions":{"view_html":"https://pith.science/pith/SNTZPZRFUXY7GWP6URUOSH65FC","download_json":"https://pith.science/pith/SNTZPZRFUXY7GWP6URUOSH65FC.json","view_paper":"https://pith.science/paper/SNTZPZRF","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2608.02064&json=true","fetch_graph":"https://pith.science/api/pith-number/SNTZPZRFUXY7GWP6URUOSH65FC/graph.json","fetch_events":"https://pith.science/api/pith-number/SNTZPZRFUXY7GWP6URUOSH65FC/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/SNTZPZRFUXY7GWP6URUOSH65FC/action/timestamp_anchor","attest_storage":"https://pith.science/pith/SNTZPZRFUXY7GWP6URUOSH65FC/action/storage_attestation","attest_author":"https://pith.science/pith/SNTZPZRFUXY7GWP6URUOSH65FC/action/author_attestation","sign_citation":"https://pith.science/pith/SNTZPZRFUXY7GWP6URUOSH65FC/action/citation_signature","submit_replication":"https://pith.science/pith/SNTZPZRFUXY7GWP6URUOSH65FC/action/replication_record"}},"created_at":"2026-08-04T02:10:46.342950+00:00","updated_at":"2026-08-04T02:10:46.342950+00:00"}